[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: [Libguestfs] [libnbd RFC PATCH] First stab at CI infrastructure

On Wed, Apr 07, 2021 at 05:49:30PM +0100, Richard W.M. Jones wrote:
On Wed, Apr 07, 2021 at 04:55:17PM +0200, Martin Kletzander wrote:
Polite ping.  Any thoughts?  Anything you'd like to change?

I didn't particularly understand any of it; but it's what we need so
looks good!

Well, any question you might have I'll do my best to answer it.  It took
me a while as well, but it is way more versatile and stable than other
CI setups.

One bug: All the new files that have been added probably need to be
added to the top-level Makefile.am EXTRA_DIST, certainly if they would
be in any way useful in the tarball.  However if they are not useful
in the tarball then the target "maintainer-check-extra-dist" (also in
Makefile.am) must be modified to ignore these files when I do
"make && make dist && make maintainer-check-extra-dist".

I did not really check more than just running the tests, but if there is
any workflow that you have, then we can incorporate it in there, so that
it checks as much as possible.  I tried running `make distcheck` and
there were some harder failures than these, so I wanted to start small.

I also looked at the results:


and choosing the top "failed" link arbitrarily takes me to:


where there are some failures.  We don't support mingw* platforms for
libnbd (we do for nbdkit), so for libnbd those platforms can
immediately be dropped.

Good to know.  I started with the list from libvirt (which is the most
complete I know) and figured we will tune it later.  I'm removing the
mingw builds then, thanks.

But there are some other non-mingw failures, again picking one
at random:


the failure is:

 go install: version is required when current directory is not in a module

which is (or was) a real bug!  I've since fixed it upstream, but it
seems to show the thing works.

As far as I know libnbd should currently build on Rawhide, so could
you pull in the updates, drop mingw*, and see if the tests start to

There are couple more, but they are the same failure.  Basically
anything that builds with clang (BSD and MacOS by default and CentOS 8
is running both GCC and clang variants) will error out due to unused
string vector functions.  Unfortunately those are generated.



GCC does not find this even with --enable-gcc-warnings (already used).

One last thing that I forgot to mention, but I definitely wanted to ask
about is whether there is a reason behind not erroring out when some
configure option is requested, but cannot be fulfilled.  For example
`--with-libxml2` when pkg-config cannot find it will just result in a
warning that most people will miss.  If new dependencies are required
and we forget to add them to the list of packages for CI, then instead
of the CI failing and us noticing pretty quickly we will just build
without the support for some part of the code and completely missing it
in the tests as well.

Anyway, I updated some details and ran it again:


Let's see what's going to pop up next.


Richard Jones, Virtualization Group, Red Hat http://people.redhat.com/~rjones
Read my programming and virtualization blog: http://rwmj.wordpress.com
Fedora Windows cross-compiler. Compile Windows programs, test, and
build Windows installers. Over 100 libraries supported.

Attachment: signature.asc
Description: PGP signature

[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]